Add a Open_Spiel Stratego implementation as well adapting MCTSAgent by defining different types of it

[BluemlJ]

With this pull request, I add an interface to CrazyAra which enables the use of my own Stratego implementation which is based on Open_Spiel. This PR adds two major extensions to CrazyAra:

First the extension of a Stratego environment:

The interface is based on a new state (engine/src/environments/stratego_related/), based on the existing interface for open_spiel games. It also adds my own fork of open_spiel as a submodule because it is not merged with the official open_spiel project yet. (.gitmodules)

To use CrazyAra for Stratego a new game mode called MODE_STRATEGO is added to all necessary files as well as the CMakeList:

• engine/CMakeLists.txt
• engine/src/agents/agent.cpp (some improved console output for better understanding Stratego moves)
• engine/src/nn/tensorrtapi.cpp
• engine/src/rl/selfplay.cpp
• engine/src/stateobj.h
• engine/src/uci/optionsuci.cpp
• engine/src/uci/variants.h

It is compatible with the Profiling modes as well as the RL mode and is tested with the TensorRT backend.
A building test was also added to the GitHub workflow as well as the possiblity to start the variants workflow out of Github to make testing easier (variants.yml).

The second extension was the addition of MCTSAgent Types. In my thesis, I tested different agent types on imperfect information games. For this, I added new agents which inherit the basic MCTSAgent

• engine/src/agents/mctsagentbatch, runs multiple mcts agents and combine the results at the end
• engine/src/agents/randomagent, plays random moves wihtout mcts
• engine/src/agents/mctsagenttruesight, removes hidden information from imperfect information games to play on a perfect infomation game

I also adapted MCTSAgent for this and added some specific handling for Stratego (like playing starting positions twice with alternating colors). I also added three new modes which are currently only tested on Stratego to run comparisons between different MCTSAgent types and models more easily, directly on the console without using additional software.

• mctsarena, gets 2 different agent types, the number of matches to play and uses the same model on both agents
• mctstournament, gets a list of agent types to play a round-robin tournament of n games between all given agents. All agents use the same model.
• tournament, gets tuples of agen types and models to play a round-robin tournament of n games between all given tuples. The models must be defined in separate folders next to the binary.

I also added a Dockerfile for Stratego which extends the current docker file by installing open_spiel and the dependencies.
(engine/src/rl/Dockerfile_Stratego)